Towards Sustainable Energy: Predictive Models for Space Heating Consumption at the European Central Bank
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsAn interesting study on Predictive Models for Space Heating Consumption at the European Central Bank seeking sustainability.
Some recommendations:
- The manuscript could provide a more detailed justification for why While XGBoost is identified as the most effective model.
- I would recommend preparing a discussion of the potential limitations of the models.
- I recommend discussing how the findings of this study could be applied to other types of buildings, such as residential. This would give the research a broader impact.
- The clarity of the Figures could be improved (please increase the text of Figures 1 and 2).
Author Response
An interesting study on Predictive Models for Space Heating Consumption at the European Central Bank seeking sustainability.
Some recommendations:
- The manuscript could provide a more detailed justification for why While XGBoost is identified as the most effective model.
Response: Thank you for your feedback. We have added a detailed justification in Sections 4.1.1 and 4.2.1, highlighting XGBoost’s ability to handle missing data, optimize decision trees through gradient boosting, capture complex nonlinear relationships, and leverage built-in regularization to enhance predictive accuracy.
- I would recommend preparing a discussion of the potential limitations of the models.
Response: Thank you for your feedback. We have included a discussion on the potential limitations of the models in Section 5, addressing dataset constraints, computational complexity, weather variability, and the need for larger datasets for deep learning models.
- I recommend discussing how the findings of this study could be applied to other types of buildings, such as residential. This would give the research a broader impact.
Response: Thank you for your suggestion. We have discussed the applicability of our findings to residential buildings in Section 6, highlighting how XGBoost can be adapted for dynamic heating patterns and energy efficiency optimization.
- The clarity of the Figures could be improved (please increase the text of Figures 1 and 2).
Response: Thank you for your suggestion. Text of figures is increased to provide more clarity.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript presents a machine learning-based predictive model for space heating consumption at the European Central Bank headquarters, focusing on sustainability and energy efficiency. The study compares various machine learning and deep learning models. Therefore, I believe there is a good chance that this manuscript can be a publication here. However, my comments below need to be very carefully addressed.
- How did the authors construct the loss function?
- Could an analysis of computational efficiency be included, comparing training times and model complexity trade-offs for real-world implementation?
- Could an extended evaluation of the models' interpretability be included, such as SHAP (Shapley Additive Explanations) values, to identify the most influential predictors?
- As the authors focus on ML-driven approach for predicting on space heating. A relevant study in this regard should be included and mentioned in the introduction in the manuscript in the proper location: Atmosphere 16 (1), 95, 2025.
- How do variations in the hyperparameter tuning of ensemble models, such as learning rate and max depth, influence predictive accuracy?
Author Response
The manuscript presents a machine learning-based predictive model for space heating consumption at the European Central Bank headquarters, focusing on sustainability and energy efficiency. The study compares various machine learning and deep learning models. Therefore, I believe there is a good chance that this manuscript can be a publication here. However, my comments below need to be very carefully addressed.
- How did the authors construct the loss function?
Response: Thank you for your comment. We have provided details on the construction of the loss function in Section 3.3, explaining the use of Mean Squared Error (MSE) as the primary optimization criterion within the XGBoost framework.
- Could an analysis of computational efficiency be included, comparing training times and model complexity trade-offs for real-world implementation?
Response: Thank you for your suggestion. Computational efficiency and model complexity trade-offs were not included in our study as the primary focus was on predictive accuracy across different models and feature sets. However, future work will explore these aspects to assess the feasibility of real-world implementation.
- Could an extended evaluation of the models' interpretability be included, such as SHAP (Shapley Additive Explanations) values, to identify the most influential predictors?
Response: Thank you for your suggestion. While our study primarily focused on evaluating predictive accuracy, we recognize the value of model interpretability. Future work will incorporate SHAP values to analyze the influence of individual features, providing deeper insights into model decision-making.
- As the authors focus on ML-driven approach for predicting on space heating. A relevant study in this regard should be included and mentioned in the introduction in the manuscript in the proper location: Atmosphere 16 (1), 95, 2025.
Response: Thank you for your suggestion. We have incorporated the relevant study from Atmosphere, Volume 16, Issue 1, 2025, in the introduction at the appropriate location
- How do variations in the hyperparameter tuning of ensemble models, such as learning rate and max depth, influence predictive accuracy?
Response: Thank you for your suggestion. We have discussed the influence of hyperparameter tuning, including learning rate and max depth, on predictive accuracy in Section 3.2.
Reviewer 3 Report
Comments and Suggestions for AuthorsThis study evaluates machine learning and deep learning models for estimating the space heating consumption of the ECB headquarters. Models such as KNN, SVR, DT, LR, XGBoost, RF, LSTM, and GRU are used in the study. The study aims to increase model accuracy by using a wide feature set, including data from indoor and local weather stations. The results show that XGBoost achieves the highest accuracy rate, and its R² value reaches 0.966, especially when trained with indoor weather data. This study compares machine learning and deep learning models in a wide range. The accuracy of the predictions is increased by using data from four weather stations inside and around the building. More accurate heating consumption predictions promote energy efficiency and green energy use. However,
1- The study only focuses on the European Central Bank building, its applicability to different building types has not been evaluated.
2- Its success is debatable since it only addresses a certain region.
3- It has been observed that deep learning models such as LSTM and GRU perform lower than ensemble models such as XGBoost.
4- Dimensionality reduction methods such as PCA or RFE were not used in the study, only correlation-based feature selection was made, PCA and RFE analysis results should be discussed.
The study covers analysis results that support sustainable energy management by providing a more accurate estimation of space heating consumption in office buildings. I believe that it will make a significant contribution to increasing model accuracy, especially by integrating indoor weather data. In this sense, I kindly recommend that the study be corrected by taking into account the kind recommendations I have stated above.
Author Response
Comments and Suggestions for Authors
This study evaluates machine learning and deep learning models for estimating the space heating consumption of the ECB headquarters. Models such as KNN, SVR, DT, LR, XGBoost, RF, LSTM, and GRU are used in the study. The study aims to increase model accuracy by using a wide feature set, including data from indoor and local weather stations. The results show that XGBoost achieves the highest accuracy rate, and its R² value reaches 0.966, especially when trained with indoor weather data. This study compares machine learning and deep learning models in a wide range. The accuracy of the predictions is increased by using data from four weather stations inside and around the building. More accurate heating consumption predictions promote energy efficiency and green energy use. However,
1- The study only focuses on the European Central Bank building, its applicability to different building types has not been evaluated.
Response: Thank you for your comment. While our study focuses on the European Central Bank building, it includes multiple scenarios capturing diverse operational and environmental conditions, enhancing its applicability. This limitation has been acknowledged in Section 5.
2- Its success is debatable since it only addresses a certain region.
Response: Thank you for your comment. While the study focuses on a specific region, the methodology and findings are broadly applicable to similar climates and building types, ensuring wider relevance. This limitation has been acknowledged in Section 5.
3- It has been observed that deep learning models such as LSTM and GRU perform lower than ensemble models such as XGBoost.
Response” Thank you for your comment. We have addressed this observation in Section 4.1.1 and 4.2.1, highlighting the lower performance of LSTM and GRU compared to XGBoost.
4- Dimensionality reduction methods such as PCA or RFE were not used in the study, only correlation-based feature selection was made, PCA and RFE analysis results should be discussed.
Response: Thank you for your comment. PCA and RFE were not applied as the study focused on correlation-based feature selection, which retains interpretability and ensures that the most relevant features are chosen based on their actual impact on SHC. Unlike PCA, which transforms features into principal components, correlation-based selection preserves the original feature meanings, making the model more explainable and aligned with real-world factors.
The study covers analysis results that support sustainable energy management by providing a more accurate estimation of space heating consumption in office buildings. I believe that it will make a significant contribution to increasing model accuracy, especially by integrating indoor weather data. In this sense, I kindly recommend that the study be corrected by taking into account the kind recommendations I have stated above.
Reviewer 4 Report
Comments and Suggestions for AuthorsThis study demonstrates organizations' capacity to develop in-house energy management
solutions, enabling them to fulfill their energy responsibilities autonomously. The EPB European
Directive mandates energy performance standards for buildings, making compliance with energy efficiency measures crucial. Despite this, research focusing on organizational energy management is limited. This study highlights the potential for organizations to leverage their resources for energy efficiency in alignment with European standards. By employing advanced models to accurately predict SHC, we offer a framework to enhance energy management practices, support sustainable energy usage, and contribute to global decarbonization efforts. pLease find comments as below;
- Critical analysis is missing in the introduction section should be strengthened while revising this section.
- Section 1 and 2 should be integrated. Authors should follow the publication standards and reuired criteria.
- The accuracy analysis is required.
- Feasture selection should be justified scientifically.
- Interpretation of the results should be improved significantly.
- The organization of the manuscript does not meet the Journal's standards.
Author Response
This study demonstrates organizations' capacity to develop in-house energy management
solutions, enabling them to fulfill their energy responsibilities autonomously. The EPB European
Directive mandates energy performance standards for buildings, making compliance with energy efficiency measures crucial. Despite this, research focusing on organizational energy management is limited. This study highlights the potential for organizations to leverage their resources for energy efficiency in alignment with European standards. By employing advanced models to accurately predict SHC, we offer a framework to enhance energy management practices, support sustainable energy usage, and contribute to global decarbonization efforts. pLease find comments as below;
- Critical analysis is missing in the introduction section should be strengthened while revising this section.
Response: Thank you for the suggestion. We have strengthened the critical analysis in Section 1 by highlighting the limitations of existing methods and the need for data-driven approaches.
- Section 1 and 2 should be integrated. Authors should follow the publication standards and required criteria.
Response: Thank you for the suggestion. We have integrated Sections 1 and 2 to align with publication standards and required criteria.
- The accuracy analysis is required.
Response: Thank you for your feedback. Since this is a regression study, accuracy is not an applicable metric. Instead, we have conducted a comprehensive evaluation using standard regression benchmarks such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R² score to ensure robust model assessment.
- Feasture selection should be justified scientifically.
Response: Thank you for your comment. We have justified the feature selection scientifically in Section 2.5, referencing relevant literature and explaining the significance of each feature.
- Interpretation of the results should be improved significantly.
Response: Thank you for your feedback. We have significantly improved the interpretation of the results in Section 4.
- The organization of the manuscript does not meet the Journal's standards.
Response: Thank you for your feedback. We have revised the manuscript to align with the journal's standards.
Reviewer 5 Report
Comments and Suggestions for AuthorsThe strenght of the article is handling predictive models for space heating consumption in high-volume buildings. Importance is comparing the precision of the different methods.
- In the middle of page 6 after Formula 1 the 8-th row is named „,this data set is localized.“. What does this expression mean?
- Tables 2 and 3 need more precise explanation. What is the meaning of set 1 and set 2?
- Please describe the methodology depicted in Figure 2 more.
- In conclusion, the 5th row from the top is written „…Feature Set 2 to make exact predictions.“ What does it mean exactly? How big is accuracy?
- The conclusion needs to show some numerical indicators: The article makes many calculations, and the conclusion does not include numerical comparisons of results.
Author Response
Comments and Suggestions for Authors
The strength of the article is handling predictive models for space heating consumption in high-volume buildings. Importance is comparing the precision of the different methods.
- In the middle of page 6 after Formula 1 the 8-th row is named „,this data set is localized.“. What does this expression mean?
Response: Thank you for your comment. We have rephrased it for clarity.
- Tables 2 and 3 need more precise explanation. What is the meaning of set 1 and set 2?
Response: Thank you for your feedback. We have provided a more precise explanation of Feature Set 1 and Feature Set 2 in Section 2.6.1.
- Please describe the methodology depicted in Figure 2 more.
Response: Thank you for your comment. We have provided a more detailed description of the methodology depicted in Figure 2 in Section 3.
- In conclusion, the 5th row from the top is written „…Feature Set 2 to make exact predictions.“ What does it mean exactly? How big is accuracy?
Response: Thank you for your comment. We have rephrased the line for clarity in Section 6
- The conclusion needs to show some numerical indicators: The article makes many calculations, and the conclusion does not include numerical comparisons of results.
Response: Thank you for the suggestion. We have incorporated numerical indicators in Section 6 to support the conclusions with quantitative results.